Optimal Noise Subtraction-Based Fault Components Extraction for Machinery Fault Diagnosis
Bingchang Hou, Dong Wang
Abstract
Fault components extraction is crucial to machinery fault diagnosis (MFD). As of today, most existing fault components extraction methods cannot provide an optimal estimation of fault components frequencies, and they are prone to be affected by interferential components. To solve these problems, an optimal noise subtraction (ONS) method is proposed in this paper. The ONS is based on an optimized weights spectrum (OWS), whose basis is the convex optimization modeling of healthy and faulty signals. Considering spectral coherence (SC) as a powerful cyclic frequency-spectral frequency diagram for exhibiting cyclo-stationary fault features, fault components enhanced by the ONS are subsequently used as an input to acquire an improved SC for machinery fault diagnosis. Experiments on real-world incipient bearing and gearbox fault signals have validated the effectiveness and superiority of the proposed ONS-based improved SC. The proposed ONS can effectively extract fault components and the improved SC diagram can exhibit clear fault signatures for MFD.